Tutorial #1: Statistical Methods for Safety Surveillance

The statistical methods used for data-mining or signal detection of drug-adverse event combinations from large drug safety databases such as FDA’s Adverse Event Reporting System (AERS, now FAERS), consisting of spontaneous reports on adverse events for post-market drugs are called passive surveillance methods. On the other hand, the statistical signal detection methods for longitudinal data, as the data accrues in time, are called active surveillance methods. A review of the most commonly used passive surveillance statistical methods, along with a likelihood ratio test (LRT) based method, recently developed by the instructors, will be discussed in detail. A live demo of the LRT tool on AERS data will be presented. Extensions of LRT for active surveillance will also be presented in detail.

The tutorial will consist of four modules. In Module-I, a review of commonly used Bayesian and Frequentist methods for signal detection in passive surveillance will be given. In Module-II, the LRT method will be discussed and a simulation study will be presented to assess the performance characteristics of LRT such as type-I error, false discovery rate, power and sensitivity. In Module-III, a live demonstration of the LRT tool to certain drugs (or drug classes) or adverse events (AEs) (or combinations of AEs) from the AERS database will be given, and the signals detected from the LRT method will be compared with the ones from some other commonly used methods. Finally, in Module-IV, extensions of the LRT methodology to a longitudinal clinical database and a brief discussion on how to handle excessive zeros in the data will be presented. 

At the conclusion of this tutorial, participants will:

Gain familiarity with data-mining in drug-safety database by learning:

• Explain commonly used methods for disproportionality analysis in drug safety database
• Demonstrate new method for disproportionality method that controls false positive and have good sensitivity and power, and a live demo of the accompanying tool
• Review extensions of newly developed method to longitudinal data from active surveillance